Semi-supervised logistic discrimination via graph-based regularization

MI: Global COE Program Education-and-Research Hub for Mathematics-for-IndustryグローバルCOEプログラム「マス･フォア･インダストリ教育研究拠点」We address the problem of constructing a nonlinear model based on both classified and unclassified data sets for classification. A semi-supervised logistic model with Gaussian basis expansions along with technique of graph-based regularization method is presented. Crucial issues in our modeling procedure are the choices of tuning parameters included in the nonlinear logistic models. In order to select these adjusted parameters, we derive model selection criteria from the viewpoints of information theory and Bayesian approach. Some numerical examples are conducted to show the effectiveness of our proposed semi-supervised modeling strategies